1. Hold Thru Rank Option
2. Access to Previous Values
3. Improved Error Detection

Hold Thru Rank is the ability to not sell a position in a backtest until it drops below a given screen rank.  For example, let’s say you develop a model to hold 3 positions.  That model will, by default, sell a position when it drops to position number 4.  Let us assume you want to reduce turnover and are satisfied that there is insignificant difference in position 3 and positions 4 and 5.  You might enter a Hold Thru Rank value of 5 to model this.  Here is a model where I did just that.

The Hold Thru Rank is entered just below the number of positions in the Backtest Entry Form.  I should note this is a Premium Access feature due to the additional processing required.

Accessing Previous Field Values is a potentially exciting addition to our screening parameters.  This allows for the testing of rising and falling RSf values, or rising and falling Moving Averages.  Many of you have requested the ability to test for rising RSf values, and this capability was added to the Price Alerts feature some time ago. It is now possible to test this setup within the screener.  Access to previous values is through the function PREVAL ( field, days ), where field is the database field such as [RSf] or [SMA-50], and days is the number of market days prior to the current date.  For example, PREVAL([RSf], 5) would return the value of RSf from 5 market days earlier.

In addition to a single field, you can also include a formula of multiple fields within this function, such as PREVAL([Rtn-1mo] + [Rtn-3mo] + [Rtn-6mo]), 10).  This would return the sum of those three values from 10 days prior.

So how is this used within the screen entry form?  While it can be used directly in a filter rule, my preference is to use it in a User Defined Variable, because the value of all UV’s are shown when you view a Current Screen Picks page.  In this example below, I define a UV called _preRSf to be the RSf value from 5-days prior.  This is a nice simple definition, 

and when the Current Picks page is viewed the previous RSf values are shown.  My goal in this screen was to view funds where the RSf value had risen by 5 points over the past 5 days.  In addition to this UV I need a filter such as [RSf] – [_preRSf] > 5, as shown here.    Unfortunately, all of the text is not visible in the first entry box, but it is [RSf] – [_preRSf].

This screen, which is also limited to World Sector Funds, can be viewed here, and as of this writing the results look like this:

Remember this is not just limited to RSf.  We could just as well be screening for funds where the 3-month return today is greater than the 3-month return one month ago, or funds where price is above some MA, but closer to the MA than it was a week ago.  Let your imagination run, but you will have to be a Premium Access user because this takes serious processing.

This is the first function we have tried to add to our backtester since moving to new server hardware a few months ago.  What other capabilities would you like to see added.  We have considered a Percentile Rank function that would allow us to have RSf like numbers based on any field or fields in the database.  Please let us know your thoughts.

We have also made some improvements in Error Detection on the Screen Entry Form.   Some of you may have noticed this recently in the UV section of the form, but we have now broadened it to the filter sections as well.  I mention it here because it is possible we could be preventing a perfectly legitimate entry, and if you believe that is the case then we need to know about it.  We have increased our monitoring of system errors recently and made several changes to improve entries, validate entries, and restate some vague error messages that were being sent back.  This is all in an effort to improve the user friendliness of the site and decrease the frustration level some users have been experiencing.  Rest assured we are not through with this process and would appreciate your input into improvements you would like to see made.

In reviewing this paper I wished they had focused on trading days pre- and post- month end rather than calendar day.  With different months having different numbers of days and weekends scattered through, it was hard to see a path from their research to a trading strategy, and a trading strategy is what we want. The concept of monthly seasonality has been studied many times before, and I reviewed a good number of the links that Google offered.  The earliest was from 1984 in a paper titled A Monthly Effect in Stock Returns, by Robert A. Ariel.  Another more recent paper is Equity Returns at the Turn of the Month by John McConnell and Wei Xu, from 2006.  These studies all focused on indexes and were longer term than one can achieve with ETFs, but ETFs provide a convenient vehicle to review recent history.

Since our ETFScreen.com site is based on return data, it was relatively easy to summarize the stats once we developed a proper numbering scheme for the days.  We decided to focus our analysis on the trading days from 12 days prior to the last day of the month, to 12 days after the last day of the month.  Since the typical month has only 21 trading days, this gives us some overlap at the extremities, so any mid-month anomalies are not terminated prematurely.  The  chart above shows the data for SPY, which is also tabulated below.  Day 0 is the last trade day of the month, and returns are from the prior day’s close. As an example you can see from the data that SPY has been slightly negative on the last trading day of the month during this time period.

Looking at the chart above, we see that beginning with day -9, there is weakness, followed by strength in days -3 to +2 or +3, and then further weakness.  From the base data, I know that SPY has a total return of 36.15% over the 12-year period of 2001-2012.  This equates to a CAGR of 2.6%, or about 0.21% per month.  If we focus in on the bars from -3 to +2, that is 6 days, we have an average return of 0.62% per month.  This implies that over the remaining 16 days SPY has had a net loss of 0.41% per month.  Trading only these six select days would have taken that 36% total return up to about 143%, for a CAGR of roughly 7.7%

Yes, I hear you. There are transaction costs and other slippage, and this would not be achievable, and you are potentially right.  Judging from the current time period, this strategy is also not without it’s volatility.  However, we should be able to improve on the returns by fund selection and possibly some refining of the entry/exit rules with some market indicators. The important take-away from this quick study is that this “turn of the month” seasonality can be significant to our returns and should probably not be ignored.

Let me add that this effect is fairly broad based across market sectors and in many international ETFs.  I could list them here, but this link allows you to explore them for yourself.  The default symbol is SPY, but to the right of the chart are convenient links to some sector funds and a few international funds.  Beyond that, enter whatever ETF symbol you want to explore.

Like I said above, the goal is a trading strategy. I invite comments about where we should go from here and how you would like to see the data presented.

If you haven’t kept up with the blog postings on this, several of the features that have been added over the past few months will be by subscription only going forward. These include: backtests back to 2002, statistics by cycle including trade by trade results, historical data views from the pop-up calendars, the in-page pop-up charts, and more.

Is anything remaining free (aka advertising supported)? Absolutely! The traditional offerings of the site will still be free for your use. The fund performance views, the portfolios, the screener with 5-year backtests, and even the price alerts. All of this will remain free, supported by our great advertisers.

We sincerely hope you want to support the site and jump on over to the sign-up page.  The monthly subscription is just $15.95, a very minimal amount.  The sign-up page will ask you to log-in before subscribing to the Premium Access service.  This way you keep your same user name, and all your portfolios and screens remain intact.  It is an easy process.

We have some great users and look forward to providing you a quality product, and we have several planned enhancements for both levels of service that should make the site even better.  Additionally, we are targeting some server upgrades that should improve site performance.  If you have specific suggestions for the site we welcome hearing them.

In closing,  we have an update for those of you wanting log-scale charting.  That feature was tested today and should make the main screener page this weekend with further roll-out thereafter.  This will not replace the current charts but will provide an alternative view.

I also want to point out a little-used feature that comes with Premium Access.  That is a custom link that is positioned above the price chart on the left margin.  Not many of you set this up when you registered for the free trial, but some of you that did have mentioned how convenient it is.  You can set this link to point to any web address you like, but the idea is to point it to, for example, Yahoo Finance charts, or StockCharts, with a link such as http://finance.yahoo.com/q?s=SPY.  Then, when you access a price chart on ETFScreen, for any symbol there will be an easily accessible link to your specified page but with the symbol SPY replaced with whatever symbol you are viewing.  The link above the price chart will look something like AAXJ at finance.yahoo.com, depending on the link you specify.  We have provided several sample links for you in the registration process but have left it open ended so you can determine how best to use this feature.  We recognize this breaks all the rules of internet marketing, where you try to keep users on your site, but if you are one of our Premium Access users we feel you deserve the best of all available tools.  Try it out!

Sign Up Here

We’ve had a nice little bounce this week, and it could be an opportunity to exit some remaining equities and buy some bonds or such.  With that in mind, I took another look at published bond screens to see if anything there might be intriguing.  I quickly focused in on two models, Bond TAA published by az2123,  and FIXED 20 EMA 30 MO 3 published by graydrake.  For my purposes I was focused on the 2-fund, 21-day, models, and that is what the links above point to.

There are differences in these two screens, but they both seem to perform well when equities do not.  Bond TAA selects from a group of 9 hand-picked funds, where Fixed 20 EMA utilizes all funds in the Fixed Income group.  Also, Bond TAA utilizes a blend of trailing return numbers, where Fixed 20 EMA uses a single 3-month return.

The question we’d like to know is how are these likely to perform over the next month?  Obviously we don’t know that, so we’ll take a look at the 2011 period, when Washington

Model: Bond TAA

Model: Fixed 20 EMA30 MO 3

struggled to extend the debt ceiling.  The first chart is of the Bond TAA model by az2123.  This model’s stats have impressed me with a low Ulcer Index and an unusually smooth equity curve.  It is worth a look.

The second model, FIXED 20 EMA 30 MO 3, by graydrake, is equally  interesting.  Whether you begin a backtest in 2002, 2007, or 2011, this model has returned higher CAGR but also higher UI numbers than the first model; however, a look at the charts shows much of this outperformance was in the August, 2011 time period.  Will that repeat again this time? At this point, none of us know.

Both of these screens are worth taking a look at when you are looking to add some bond funds to your portfolio.  The Bond TAA screen has been the more consistent, but FIXED 20 EMA 30 MO 3 is definitely worth watching.

So what are you using to select your bond funds?  Tell us about it, or publish your favorite screen.

Chart from new Trend Explorer

We’ll start with the boring things such as new features. Few topics are more boring than statistics, so our first new feature is a volaitlity measure, or standard deviation to be precise.  These fields were actually added to the screener a few weeks ago, and a few of you have already found them under the new Volatility section of the drop-down menus.  These fields will carry the abbreviation StDv-xxx where xxx is a usual time designation.  The durations we currently measure range from 10-days to 1-year, though the resulting values are all annualized. By annualizing the numbers, it is straight forward to compare short term volatility to longer term volatility.  However, one might want to convert these values (or the return values) if they are to be used for risk adjusting returns.

Let’s move on to a new section added to the screener called Market Timing.  We recognize this topic sparks a lot of emotion (pro and con) and considered renaming it something like Macro Trend, but decided we might as well call it what it is.  This section of the screener is totally optional but allows you to specify a symbol and a rule that must equate to true before a trade is entered. As in the example shown, you could specify that the Price of SPY is greater than the 200-day Moving Average.  Whenever the price drops below the SMA-200 on a trade date, no trades will be entered.  Note this is only interpreted on the periodic trade dates of the model and will not result in a trade being terminated early because this rule triggers.  Several of you have asked for this and from running a few test screens it could prove to be beneficial.

Market Timing is a great segue into a discussion of our new screener which we have titled Trend Explorer.  Trend Explorer had its roots in the work I did last year for the post When the Defense Leads.  In this post I explored how market performance lagged when the RSf of Consumer Staples(XLP) and Healthcare(XLY) both exceed that of the broad market (SPY). Trend Explorer basically gives you a set of tools to explore similar relationships. The image shown here represents the code for this model, and the resulting chart is at the top of the page.  Whether the knowledge gained is used for market timing a portfolio or to trade into and out of an investment vehicle like TNA or TZA, the option is yours.  In addition to pulling single values as shown here, you can also SUM values or COUNT entries for a group of symbols.  These fields allow you to test scenarios like when the average RSI of a group of funds is above x, or when over 50% of the funds have a price greater than the 50-day SMA.  There is more explanation in this Trend Explorer post.

Finally, I want to provide an update  on our Premium Access service that opens up a number of features including significantly more data for backtesting.  For the next few days we are offering free Premium Access by following this link.  We realize there has been a delay on the subscription side, but for the next few days it is free.  So enjoy it!

 Note:  The Traditional Screener defaults to 5 years of data for a backtest.  To utilize the additional data available via Premium Access enter a beginning date in the “Backtest” section at the bottom of the screen entry form.  The Trend Explorer model by default utilizes the full time period for Premium Access users and 5-years of data one week delayed for non Premium Access users. 

Above is the basic entry form for the Trend Explorer model.  The only required entries are one variable definition and one entry rule.  For example, click in the middle of the form where it says “One Variable Definition Req’d”, and a pop-up will lead you through defining your first field.  You can leave the Name as the default “v1″.  Select VALUE from the drop down and a couple of other fields will show up.  Click the red arrow to the right to select [RSf] from the drop-down, and enter a symbol like SPY.  Click OK, and you have now defined a field that is defined as the value of [RSf] for SPY or some other symbol.  Before we leave this pop-up, note that there is a short description at the bottom to help you with the form.  This changes depending on what you select form the ‘Function’ dropdown.

After defining one or more variables we need to define at least an entry rule.  For our example, lets enter the entry rule v1 > 50 and click the button to run the model, see it here.  If you looked, you’ll see it was not the best model in the world, but it was easy to enter for a first model, and that was the point.  The exit rule is optional, and the assumption is made if you do not enter an exit rule, the model will exit whenever the entry rule is false.

Last year I posted about a model that exited the market when the RSf of both XLP and XLV (that’s Consumer Discretionary and Healthcare, respectively) are higher than the RSf for SPY.  That is, the model was in the market when SPY had a higher RSf than either one or both of the other two funds.  That post was titled When the Defense Leads and code for that model is shown here.   For the past 5-years it has shown a CAGR of about 3.7% vs. 0.4% for the SPY, with a third the drawdown.  Not all bad for trading the SPY only.

Utilizing some of the other capabilities of our Trend Exporer, we can build on this concept.  SPY is not not known to imply bullish sentiment, it represents the broad market. Likewise, XLP and XLV are not the only funds that imply a defensive sentiment.  Let us hypothesize that when the bullish funds Industrials (XLI), Technology(XLK), and Consumer Discretionary(XLY) are stronger than the defensive funds Utilities(XLU), Healthcare(XLV), and Consumer Staples(XLP), it is a good time to be invested. The code in the box shows two symbol groups defined.  If you view the screen, you will see I defined my symbol groups like above and then defined two variables that represent the sum of the RSf of the funds in each group.  Since there are three funds in each group, and they all existed for the entire time period, I just compared the sums rather than calculating averages, which I’ll show in a later example.  By utilizing the three symbols, our CAGR over this time period increased to 7.8% from 3.7% in our prior example.  This is a good time, however, for a word of caution.  ETFs are relatively new and do not offer a tremendous amount of history to backtest, and the 5 years tested here is minimal.  In fact, if you begin these two tests in 2002, like our Premium Access users can do, you will find much closer results.  Therefore, be cautious in both the development and the interpretation of any models.  That said, let’s see what else we can do.

This model demonstrates averaging of values and entering and exiting using different rules.  Basically we define a group of symbols, in this case a group that is highly correlated with IWM, the Russell Small Cap fund.  The model calculates the average 2-day RSI for these funds and enters on perceived pull-backs(< 15) and exits on a return to mid-range(> 60).  You will note that we count the number of funds as a variable instead of manually counting how many symbols are in our list.  This is because some funds in our list may not have traded over the full history being tested, and by utilizing the COUNT feature we get a valid count and therefore a valid average.  While we have an AVG function in the works, we currently must use a combination of SUM and COUNT and then calculate the average in the Entry and Exit rules.  This is a bit more cumbersome than we like, but it works.  Like the other models mentioned, this model is shown more for demonstration purposes than because we think it has any trading value.

We are very excited by the possibilities this new screener offers and expect many of you will find it beneficial.  One early use has been to identify a handful of different models that together could be used to judge market direction.  Other uses have focused on trading a particular leveraged vehicle.  Whatever your personal use, please note the limitations mentioned earlier and be cautious in your interpretation because backtest results over a short period can be misleading.  May the trend be with you!

Why:  There are numerous reasons for this.  First, those funds aren’t going to be touched by many of you anyway, and they are a nuisance to filter through.  Besides that, the number of these funds has grown through the years with the saturation of the ETF market with niche funds.  This has become obvious as we have added data to our database and run some test screens against it.  What has also become obvious is that some screen results benefit from these funds being included when in actuality they are not tradeable.  We feel it is in everyone’s best interest to make some type of change.

Another issue this would eliminate is the difference in screen results intra-day and after the close.  With the current methodology the extremely low-volume funds are not updated intra-day so they do not show up in screen results.   After the close they are pulled in and can change screen results and backtest results as well.  Our proposal would be to use the same filter for intra-day updates as is used in the screener.

The common approach would be to use a volume based filter, but after much study we found there are several low volume but consistently traded funds that might could be traded by a small trader.  Inversely, we found there are several funds that show relatively decent average volume but trade sporadically.  They just seem to trade large volumes when they trade.

We are considering making this change soon but want to give you, our users, an opportunity to comment on the direction we are headed.  Please let us know your thoughts.

Several have inquired as to the additional data being added to our backtesting application and the inclusion of a premium subscription service. We’re targeting a couple of weeks from now and are adding data back to at least 2002 and probably 2001.  This will provide looks into several differing market types, but users will need to watch the funds they include in their screens because there weren’t that many ETFs in 2001 ( less than 100 on Jan 1 of that year ).  We should also remind everyone that the current 5-years of data will remain subscription-free after these changes.

Another ‘usability’ enhancement we have lined up as part of our Premium Access package is in-page charts.  These are still being tweaked, but you can now see them in action on your portfolios.  Personally, I find it a much faster way to quickly review my various portfolios.  If you don’t have a portfolio set up you can try it out on my Euro Players portfolio.  Look for the image by the fund names.

Please share your thoughts.

Additionally, we are releasing the capability to view any Portfolio or Screen at a date in the past.  Anywhere you see a calendar like the one to the right, just click it and select a date.  This image is from a portfolio, but the calendar is also on all screen picks pages just above the data table.  Soon it will be making its way onto other data tables throughout the site.  So, if you’ve ever wondered what a screen looked like just before the latest correction, now you can find out.  We are kicking this feature off with a year’s worth of historical data, but that, too, will be supplemented in the near future.

As was mentioned above, we are also making trade data available to supplement backtest results.  A new button will now appear near the bottom of the backtest statistics box titled View Results By Cycle.  This button will lead to the first of two new pages, this one showing results for each cycle.  On that page you can see how the backtest results vary with start date.  Some screens are fairly consistent, and some appear almost random.  Click the cycle number in the table header, and you will go to that cycle’s page which includes a performance chart, some relative statistics, and period by period trade data.  These features were mostly covered in the More on a Rolling Starts Backtest post, so you can see it there, or just check out your favorite backtest.   Many of you have asked for this data for a while, and we are glad to be able to make it available.

These two new features are currently available for anyone who is registered on the site and logged in.  They are part of a package of features that will be available by a small subscription fee starting in a few weeks.  Since these features are ready and have been in heavy demand, we wanted to let our current users have access while the remainder of the package is assembled.  Enjoy!

This article will take a more in-depth look at the numbers behind the averages.  For this I chose a published screen by ‘inconversable’ titled CDGR – Amer. C. F. Mk 5.  This screen holds 3 positions and rebalances every 5 days.  As you can see from this image, the CAGR results are shown as 27.9% versus -1.1% for the SPY.

So what is behind this number?  Actually, there are 5 independent backtests behind these results.  Those 5 backtests start on 5 consecutive days beginning on 5/21/2007, and all end on 5/18/2012.  Did they all produce the exact same statistics?  Of course not.  There is a large amount of random noise in security prices so it is not unusual for these independent cycles to result in wildly different results.  The table below illustrates the 5 different cycles for this model.

The top couple of lines of this table show the first and last rebalance dates for each cycle.  You can see these are on 5 consecutive dates but might note that cycle 1 appears out of order.  This means nothing for these results; but, these tables will become more commonplace on the site, and the cycles are defined so cycle 1 on a 5 day model is always the same cycle.  That way, if you are following a model you can pick a cycle number and have real results to compare against.

Let’s refocus on the numbers, particularly on the CAGR line.  You can see that these cycles ranged from a 16% CAGR all the way up to 33%.  If you had studied backtest results based on a single cycle, you could have seen any of these numbers and possibly made an incorrect decision about the screen. Backtests are rarely a perfect indicator of future results, but backtests that utilize as much data as possible statistically improve the odds.

Similar variances exist in the other statistics as well.  In fact, if you look at the SPY results you will see a small variation there caused by the different starting dates.  Depending on market dynamics, there have been a few times that SPY returns were significantly different for models of different holding periods measured over the same time period.  Although rare, it has caused a few of you to rightfully question the numbers.

Going a step further, we can look at any cycle for actual trade-by-trade results grouped by date.  The table to the left shows the last few trades for cycle 1 of the screen above.

We are in the final stages of testing  before releasing these pages for your usage.  We have several other enhancements on deck as well.  To support these features it took a good bit of investment in our backtesting server/software, and there is a lot more data to pass around.  We are confident you will find this increased documentation valuable and supportive of your decision making.

Our system is set up where you have a screen definition and optionally may have multiple backtests run against that screen. Once you name a Favorite Screen, that name applies to the screen and all related backtests. You have the option to flag the screen (current picks) as the favorite or one or more backtests. I’ve found it more useful to flag the backtest that I want to follow than to flag the Current Screen Picks, but your usage and experience may differ.

So what about those Published Screens? Published Screens are a great way for you to share some of your favorite screens with other users. This site is all about filtering down the universe of ETFs to the few we think are investment worthy. We all try various screen definitions to see how they have performed. Many of us duplicating efforts with one another. By publishing a screen you think has potential you present it to the group and open it up to critique. Users can rate your screen with a star rating system and can enter into a discussion about your screen. Hopefully such will lead to even better screens. Published Screens let us leverage the combined knowledge of our user base.

I’ve released this capability with a couple of basic screens published. My hopes are that you will check out what has been published, rate them, comment on them, and publish some screens of your own.

And to tie this all together, you can always flag a Published Screen as a Favorite so you can easily find it later.

p.s. In a related move, the Screener now supports symbol lists without the need to create a portfolio. That makes life easier.

Go to the Screener and get publishing!